At the International Electronic Devices Meeting (IEDM) Intel provided details about its 18A (1.8nm-class) process, while TSMC's showed off its N2 (2nm-class)
Those on the ground think that Intel’s offering could offer superior performance, while TSMC’s process may provide higher transistor density.
According to analysis from TechInsights, TSMC's N2 process boasts a high-density (HD) standard-cell transistor density of 313 million transistors per square millimetre (MTr/mm²), significantly surpassing Intel's 18A, which is rated at 238 MTr/mm², and Samsung's SF2/SF3P, which comes in at 231 MTr/mm².
These figures only pertain to HD standard cells. Modern high-performance processors typically incorporate a mix of HD, high-performance (HP), and low-power (LP) standard cells, along with additional optimisations such as TSMC's FinFlex and NanoFlex.
How Intel's and TSMC's HP and LP standard cells compare directly remains unclear. While TSMC appears to have a lead in transistor density, the gap may not be as pronounced across all standard-cell types.
Both Intel and TSMC presented papers at IEDM outlining the advantages of their respective manufacturing processes. Still, there is no definitive method for a direct, like-for-like comparison of the two technologies.
On the performance front, TechInsights suggests that Intel's 18A process may hold an advantage over TSMC's N2 and Samsung's SF2. However, this assessment is based on a methodology that extrapolates performance gains from past node transitions using TSMC's N16FF and Samsung's 14nm processes as baselines, which may introduce a degree of uncertainty.
Intel's 18A is expected to prioritise performance and power efficiency rather than maximising transistor density. Notably, it incorporates PowerVia, a backside power delivery network that could confer advantages in both performance and density over TSMC's N2, which lacks this feature. However, not all chips manufactured on 18A will necessarily make use of PowerVia.
TechInsights analysts suggest that chips produced on TSMC's N2 process will likely consume less power than those based on Samsung's SF2, reflecting TSMC's historical leadership in power efficiency. The extent to which Intel's 18A will compete in this regard remains to be seen, though it is expected to provide some benefits.
Intel's 18A is scheduled to enter mass production in mid-2025, coinciding with its Core Ultra 3-series 'Panther Lake' processors, which are due to be released later this year. In contrast, TSMC's N2 process is not expected to enter high-volume manufacturing until late 2025, with the first commercial products arriving no earlier than mid-2026 and broader market adoption anticipated by autumn 2026.
Samsung has yet to specify a precise timeline for SF2's high-volume manufacturing, only indicating a launch at some point in 2025.
